NATHANNematode counting for Scientia Terrae
Oktober 2024 - November 2024
Scientia Terrae conducts research and development (R&D) for companies like DCM. As part of their research, they need a computer vision model that can count nematodes in a soil sample. Nematodes are microscopic worms that cause damage to plants.
So by counting nematodes, they can scientifically demonstrate that certain products are effective in controlling these worms.
The first step in this project was to improve the dataset using data augmentation so that the model will make accurate predictions in more conditions. After this, I trained the YOLOv11 model and used state-of-the-art techniques such as SAHI, which further improved our results.
Finally, I fine-tuned the model so that the balance between recall and precision was optimal for this application.
